PDMAsyncCompletionFile.cpp@ 57358

最後變更在這個檔案從57358是 57358,由 vboxsync 提交於 10 年前
*: scm cleanup run.
屬性 svn:eol-style 設為 `native` 屬性 svn:keywords 設為 `Author Date Id Revision`
檔案大小: 46.9 KB

行
1	/* $Id: PDMAsyncCompletionFile.cpp 57358 2015-08-14 15:16:38Z vboxsync $ */
2	/** @file
3	* PDM Async I/O - Transport data asynchronous in R3 using EMT.
4	*/
5
6	/*
7	* Copyright (C) 2006-2015 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.alldomusa.eu.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_PDM_ASYNC_COMPLETION
23	#include "PDMInternal.h"
24	#include <VBox/vmm/pdm.h>
25	#include <VBox/vmm/mm.h>
26	#include <VBox/vmm/vm.h>
27	#include <VBox/err.h>
28	#include <VBox/log.h>
29	#include <VBox/dbg.h>
30	#include <VBox/vmm/uvm.h>
31	#include <VBox/vmm/tm.h>
32
33	#include <iprt/asm.h>
34	#include <iprt/assert.h>
35	#include <iprt/critsect.h>
36	#include <iprt/env.h>
37	#include <iprt/file.h>
38	#include <iprt/mem.h>
39	#include <iprt/semaphore.h>
40	#include <iprt/string.h>
41	#include <iprt/thread.h>
42	#include <iprt/path.h>
43	#include <iprt/rand.h>
44
45	#include "PDMAsyncCompletionFileInternal.h"
46
47
48	/*********************************************************************************************************************************
49	* Internal Functions *
50	*********************************************************************************************************************************/
51	#ifdef VBOX_WITH_DEBUGGER
52	static FNDBGCCMD pdmacEpFileErrorInject;
53	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
54	static FNDBGCCMD pdmacEpFileDelayInject;
55	# endif
56	#endif
57
58
59	/*********************************************************************************************************************************
60	* Global Variables *
61	*********************************************************************************************************************************/
62	#ifdef VBOX_WITH_DEBUGGER
63	static const DBGCVARDESC g_aInjectErrorArgs[] =
64	{
65	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
66	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write/read." },
67	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
68	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "errcode", "VBox status code." },
69	};
70
71	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
72	static const DBGCVARDESC g_aInjectDelayArgs[] =
73	{
74	/* cTimesMin, cTimesMax, enmCategory, fFlags, pszName, pszDescription */
75	{ 1, 1, DBGCVAR_CAT_STRING, 0, "direction", "write\|read\|flush\|any." },
76	{ 1, 1, DBGCVAR_CAT_STRING, 0, "filename", "Filename." },
77	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "delay", "Delay in milliseconds." },
78	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "jitter", "Jitter of the delay." },
79	{ 1, 1, DBGCVAR_CAT_NUMBER, 0, "reqs", "Number of requests to delay." }
80
81	};
82	# endif
83
84	/** Command descriptors. */
85	static const DBGCCMD g_aCmds[] =
86	{
87	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax,.pszDescription */
88	{ "injecterror", 3, 3, &g_aInjectErrorArgs[0], 3, 0, pdmacEpFileErrorInject, "", "Inject error into I/O subsystem." }
89	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
90	,{ "injectdelay", 3, 5, &g_aInjectDelayArgs[0], RT_ELEMENTS(g_aInjectDelayArgs), 0, pdmacEpFileDelayInject, "", "Inject a delay of a request." }
91	# endif
92	};
93	#endif
94
95
96	/**
97	* Frees a task.
98	*
99	* @returns nothing.
100	* @param pEndpoint Pointer to the endpoint the segment was for.
101	* @param pTask The task to free.
102	*/
103	void pdmacFileTaskFree(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
104	{
105	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
106
107	LogFlowFunc((": pEndpoint=%p pTask=%p\n", pEndpoint, pTask));
108
109	/* Try the per endpoint cache first. */
110	if (pEndpoint->cTasksCached < pEpClass->cTasksCacheMax)
111	{
112	/* Add it to the list. */
113	pEndpoint->pTasksFreeTail->pNext = pTask;
114	pEndpoint->pTasksFreeTail = pTask;
115	ASMAtomicIncU32(&pEndpoint->cTasksCached);
116	}
117	else
118	{
119	Log(("Freeing task %p because all caches are full\n", pTask));
120	MMR3HeapFree(pTask);
121	}
122	}
123
124	/**
125	* Allocates a task segment
126	*
127	* @returns Pointer to the new task segment or NULL
128	* @param pEndpoint Pointer to the endpoint
129	*/
130	PPDMACTASKFILE pdmacFileTaskAlloc(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
131	{
132	PPDMACTASKFILE pTask = NULL;
133
134	/* Try the small per endpoint cache first. */
135	if (pEndpoint->pTasksFreeHead == pEndpoint->pTasksFreeTail)
136	{
137	/* Try the bigger endpoint class cache. */
138	PPDMASYNCCOMPLETIONEPCLASSFILE pEndpointClass = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->Core.pEpClass;
139
140	/*
141	* Allocate completely new.
142	* If this fails we return NULL.
143	*/
144	int rc = MMR3HeapAllocZEx(pEndpointClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
145	sizeof(PDMACTASKFILE),
146	(void **)&pTask);
147	if (RT_FAILURE(rc))
148	pTask = NULL;
149
150	LogFlow(("Allocated task %p\n", pTask));
151	}
152	else
153	{
154	/* Grab a free task from the head. */
155	AssertMsg(pEndpoint->cTasksCached > 0, ("No tasks cached but list contains more than one element\n"));
156
157	pTask = pEndpoint->pTasksFreeHead;
158	pEndpoint->pTasksFreeHead = pTask->pNext;
159	ASMAtomicDecU32(&pEndpoint->cTasksCached);
160	}
161
162	pTask->pNext = NULL;
163
164	return pTask;
165	}
166
167	PPDMACTASKFILE pdmacFileEpGetNewTasks(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
168	{
169	/*
170	* Get pending tasks.
171	*/
172	PPDMACTASKFILE pTasks = ASMAtomicXchgPtrT(&pEndpoint->pTasksNewHead, NULL, PPDMACTASKFILE);
173
174	/* Reverse the list to process in FIFO order. */
175	if (pTasks)
176	{
177	PPDMACTASKFILE pTask = pTasks;
178
179	pTasks = NULL;
180
181	while (pTask)
182	{
183	PPDMACTASKFILE pCur = pTask;
184	pTask = pTask->pNext;
185	pCur->pNext = pTasks;
186	pTasks = pCur;
187	}
188	}
189
190	return pTasks;
191	}
192
193	static void pdmacFileAioMgrWakeup(PPDMACEPFILEMGR pAioMgr)
194	{
195	bool fWokenUp = ASMAtomicXchgBool(&pAioMgr->fWokenUp, true);
196	if (!fWokenUp)
197	{
198	bool fWaitingEventSem = ASMAtomicReadBool(&pAioMgr->fWaitingEventSem);
199	if (fWaitingEventSem)
200	{
201	int rc = RTSemEventSignal(pAioMgr->EventSem);
202	AssertRC(rc);
203	}
204	}
205	}
206
207	static int pdmacFileAioMgrWaitForBlockingEvent(PPDMACEPFILEMGR pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT enmEvent)
208	{
209	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, enmEvent);
210	Assert(!pAioMgr->fBlockingEventPending);
211	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, true);
212
213	/* Wakeup the async I/O manager */
214	pdmacFileAioMgrWakeup(pAioMgr);
215
216	/* Wait for completion. */
217	int rc = RTSemEventWait(pAioMgr->EventSemBlock, RT_INDEFINITE_WAIT);
218	AssertRC(rc);
219
220	ASMAtomicXchgBool(&pAioMgr->fBlockingEventPending, false);
221	ASMAtomicWriteU32((volatile uint32_t *)&pAioMgr->enmBlockingEvent, PDMACEPFILEAIOMGRBLOCKINGEVENT_INVALID);
222
223	return rc;
224	}
225
226	int pdmacFileAioMgrAddEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
227	{
228	LogFlowFunc(("pAioMgr=%#p pEndpoint=%#p{%s}\n", pAioMgr, pEndpoint, pEndpoint->Core.pszUri));
229
230	/* Update the assigned I/O manager. */
231	ASMAtomicWritePtr(&pEndpoint->pAioMgr, pAioMgr);
232
233	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
234	AssertRCReturn(rc, rc);
235
236	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint, pEndpoint);
237	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_ADD_ENDPOINT);
238	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.AddEndpoint.pEndpoint);
239
240	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
241
242	return rc;
243	}
244
245	#ifdef SOME_UNUSED_FUNCTION
246	static int pdmacFileAioMgrRemoveEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
247	{
248	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
249	AssertRCReturn(rc, rc);
250
251	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint, pEndpoint);
252	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_REMOVE_ENDPOINT);
253	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.RemoveEndpoint.pEndpoint);
254
255	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
256
257	return rc;
258	}
259	#endif
260
261	static int pdmacFileAioMgrCloseEndpoint(PPDMACEPFILEMGR pAioMgr, PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint)
262	{
263	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
264	AssertRCReturn(rc, rc);
265
266	ASMAtomicWritePtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint, pEndpoint);
267	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_CLOSE_ENDPOINT);
268	ASMAtomicWriteNullPtr(&pAioMgr->BlockingEventData.CloseEndpoint.pEndpoint);
269
270	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
271
272	return rc;
273	}
274
275	static int pdmacFileAioMgrShutdown(PPDMACEPFILEMGR pAioMgr)
276	{
277	int rc = RTCritSectEnter(&pAioMgr->CritSectBlockingEvent);
278	AssertRCReturn(rc, rc);
279
280	rc = pdmacFileAioMgrWaitForBlockingEvent(pAioMgr, PDMACEPFILEAIOMGRBLOCKINGEVENT_SHUTDOWN);
281
282	RTCritSectLeave(&pAioMgr->CritSectBlockingEvent);
283
284	return rc;
285	}
286
287	int pdmacFileEpAddTask(PPDMASYNCCOMPLETIONENDPOINTFILE pEndpoint, PPDMACTASKFILE pTask)
288	{
289	PPDMACTASKFILE pNext;
290	do
291	{
292	pNext = pEndpoint->pTasksNewHead;
293	pTask->pNext = pNext;
294	} while (!ASMAtomicCmpXchgPtr(&pEndpoint->pTasksNewHead, pTask, pNext));
295
296	pdmacFileAioMgrWakeup(ASMAtomicReadPtrT(&pEndpoint->pAioMgr, PPDMACEPFILEMGR));
297
298	return VINF_SUCCESS;
299	}
300
301	void pdmacFileEpTaskCompleted(PPDMACTASKFILE pTask, void *pvUser, int rc)
302	{
303	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pvUser;
304
305	LogFlowFunc(("pTask=%#p pvUser=%#p rc=%Rrc\n", pTask, pvUser, rc));
306
307	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_FLUSH)
308	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, rc, true);
309	else
310	{
311	Assert((uint32_t)pTask->DataSeg.cbSeg == pTask->DataSeg.cbSeg && (int32_t)pTask->DataSeg.cbSeg >= 0);
312	uint32_t uOld = ASMAtomicSubS32(&pTaskFile->cbTransferLeft, (int32_t)pTask->DataSeg.cbSeg);
313
314	/* The first error will be returned. */
315	if (RT_FAILURE(rc))
316	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
317	#ifdef VBOX_WITH_DEBUGGER
318	else
319	{
320	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
321
322	/* Overwrite with injected error code. */
323	if (pTask->enmTransferType == PDMACTASKFILETRANSFER_READ)
324	rc = ASMAtomicXchgS32(&pEpFile->rcReqRead, VINF_SUCCESS);
325	else
326	rc = ASMAtomicXchgS32(&pEpFile->rcReqWrite, VINF_SUCCESS);
327
328	if (RT_FAILURE(rc))
329	ASMAtomicCmpXchgS32(&pTaskFile->rc, rc, VINF_SUCCESS);
330	}
331	#endif
332
333	if (!(uOld - pTask->DataSeg.cbSeg)
334	&& !ASMAtomicXchgBool(&pTaskFile->fCompleted, true))
335	{
336	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
337	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pTaskFile->Core.pEndpoint;
338	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEpFile->Core.pEpClass;
339
340	/* Check if we should delay completion of the request. */
341	if ( ASMAtomicReadU32(&pEpFile->msDelay) > 0
342	&& ASMAtomicReadU32(&pEpFile->cReqsDelay) > 0)
343	{
344	uint64_t tsDelay = pEpFile->msDelay;
345
346	if (pEpFile->msJitter)
347	tsDelay = (RTRandU32() % 100) > 50 ? pEpFile->msDelay + (RTRandU32() % pEpFile->msJitter)
348	: pEpFile->msDelay - (RTRandU32() % pEpFile->msJitter);
349	ASMAtomicDecU32(&pEpFile->cReqsDelay);
350
351	/* Arm the delay. */
352	pTaskFile->tsDelayEnd = RTTimeProgramMilliTS() + tsDelay;
353
354	/* Append to the list. */
355	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
356	do
357	{
358	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
359	pTaskFile->pDelayedNext = pHead;
360	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTaskFile, pHead));
361
362	if (tsDelay < pEpClassFile->cMilliesNext)
363	{
364	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, tsDelay);
365	TMTimerSetMillies(pEpClassFile->pTimer, tsDelay);
366	}
367
368	LogRel(("AIOMgr: Delaying request %#p for %u ms\n", pTaskFile, tsDelay));
369	}
370	else
371	#endif
372	pdmR3AsyncCompletionCompleteTask(&pTaskFile->Core, pTaskFile->rc, true);
373	}
374	}
375	}
376
377	DECLINLINE(void) pdmacFileEpTaskInit(PPDMASYNCCOMPLETIONTASK pTask, size_t cbTransfer)
378	{
379	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
380
381	Assert((uint32_t)cbTransfer == cbTransfer && (int32_t)cbTransfer >= 0);
382	ASMAtomicWriteS32(&pTaskFile->cbTransferLeft, (int32_t)cbTransfer);
383	ASMAtomicWriteBool(&pTaskFile->fCompleted, false);
384	ASMAtomicWriteS32(&pTaskFile->rc, VINF_SUCCESS);
385	}
386
387	int pdmacFileEpTaskInitiate(PPDMASYNCCOMPLETIONTASK pTask,
388	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
389	PCRTSGSEG paSegments, size_t cSegments,
390	size_t cbTransfer, PDMACTASKFILETRANSFER enmTransfer)
391	{
392	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
393	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
394
395	Assert( (enmTransfer == PDMACTASKFILETRANSFER_READ)
396	\|\| (enmTransfer == PDMACTASKFILETRANSFER_WRITE));
397
398	for (size_t i = 0; i < cSegments; i++)
399	{
400	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
401	AssertPtr(pIoTask);
402
403	pIoTask->pEndpoint = pEpFile;
404	pIoTask->enmTransferType = enmTransfer;
405	pIoTask->Off = off;
406	pIoTask->DataSeg.cbSeg = paSegments[i].cbSeg;
407	pIoTask->DataSeg.pvSeg = paSegments[i].pvSeg;
408	pIoTask->pvUser = pTaskFile;
409	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
410
411	/* Send it off to the I/O manager. */
412	pdmacFileEpAddTask(pEpFile, pIoTask);
413	off += paSegments[i].cbSeg;
414	cbTransfer -= paSegments[i].cbSeg;
415	}
416
417	AssertMsg(!cbTransfer, ("Incomplete transfer %u bytes left\n", cbTransfer));
418
419	return VINF_AIO_TASK_PENDING;
420	}
421
422	/**
423	* Creates a new async I/O manager.
424	*
425	* @returns VBox status code.
426	* @param pEpClass Pointer to the endpoint class data.
427	* @param ppAioMgr Where to store the pointer to the new async I/O manager on success.
428	* @param enmMgrType Wanted manager type - can be overwritten by the global override.
429	*/
430	int pdmacFileAioMgrCreate(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClass, PPPDMACEPFILEMGR ppAioMgr,
431	PDMACEPFILEMGRTYPE enmMgrType)
432	{
433	LogFlowFunc((": Entered\n"));
434
435	PPDMACEPFILEMGR pAioMgrNew;
436	int rc = MMR3HeapAllocZEx(pEpClass->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION, sizeof(PDMACEPFILEMGR), (void **)&pAioMgrNew);
437	if (RT_SUCCESS(rc))
438	{
439	if (enmMgrType < pEpClass->enmMgrTypeOverride)
440	pAioMgrNew->enmMgrType = enmMgrType;
441	else
442	pAioMgrNew->enmMgrType = pEpClass->enmMgrTypeOverride;
443
444	pAioMgrNew->msBwLimitExpired = RT_INDEFINITE_WAIT;
445
446	rc = RTSemEventCreate(&pAioMgrNew->EventSem);
447	if (RT_SUCCESS(rc))
448	{
449	rc = RTSemEventCreate(&pAioMgrNew->EventSemBlock);
450	if (RT_SUCCESS(rc))
451	{
452	rc = RTCritSectInit(&pAioMgrNew->CritSectBlockingEvent);
453	if (RT_SUCCESS(rc))
454	{
455	/* Init the rest of the manager. */
456	if (pAioMgrNew->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
457	rc = pdmacFileAioMgrNormalInit(pAioMgrNew);
458
459	if (RT_SUCCESS(rc))
460	{
461	pAioMgrNew->enmState = PDMACEPFILEMGRSTATE_RUNNING;
462
463	rc = RTThreadCreateF(&pAioMgrNew->Thread,
464	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
465	? pdmacFileAioMgrFailsafe
466	: pdmacFileAioMgrNormal,
467	pAioMgrNew,
468	0,
469	RTTHREADTYPE_IO,
470	0,
471	"AioMgr%d-%s", pEpClass->cAioMgrs,
472	pAioMgrNew->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE
473	? "F"
474	: "N");
475	if (RT_SUCCESS(rc))
476	{
477	/* Link it into the list. */
478	RTCritSectEnter(&pEpClass->CritSect);
479	pAioMgrNew->pNext = pEpClass->pAioMgrHead;
480	if (pEpClass->pAioMgrHead)
481	pEpClass->pAioMgrHead->pPrev = pAioMgrNew;
482	pEpClass->pAioMgrHead = pAioMgrNew;
483	pEpClass->cAioMgrs++;
484	RTCritSectLeave(&pEpClass->CritSect);
485
486	*ppAioMgr = pAioMgrNew;
487
488	Log(("PDMAC: Successfully created new file AIO Mgr {%s}\n", RTThreadGetName(pAioMgrNew->Thread)));
489	return VINF_SUCCESS;
490	}
491	pdmacFileAioMgrNormalDestroy(pAioMgrNew);
492	}
493	RTCritSectDelete(&pAioMgrNew->CritSectBlockingEvent);
494	}
495	RTSemEventDestroy(pAioMgrNew->EventSem);
496	}
497	RTSemEventDestroy(pAioMgrNew->EventSemBlock);
498	}
499	MMR3HeapFree(pAioMgrNew);
500	}
501
502	LogFlowFunc((": Leave rc=%Rrc\n", rc));
503
504	return rc;
505	}
506
507	/**
508	* Destroys a async I/O manager.
509	*
510	* @returns nothing.
511	* @param pAioMgr The async I/O manager to destroy.
512	*/
513	static void pdmacFileAioMgrDestroy(PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile, PPDMACEPFILEMGR pAioMgr)
514	{
515	int rc = pdmacFileAioMgrShutdown(pAioMgr);
516	AssertRC(rc);
517
518	/* Unlink from the list. */
519	rc = RTCritSectEnter(&pEpClassFile->CritSect);
520	AssertRC(rc);
521
522	PPDMACEPFILEMGR pPrev = pAioMgr->pPrev;
523	PPDMACEPFILEMGR pNext = pAioMgr->pNext;
524
525	if (pPrev)
526	pPrev->pNext = pNext;
527	else
528	pEpClassFile->pAioMgrHead = pNext;
529
530	if (pNext)
531	pNext->pPrev = pPrev;
532
533	pEpClassFile->cAioMgrs--;
534	rc = RTCritSectLeave(&pEpClassFile->CritSect);
535	AssertRC(rc);
536
537	/* Free the resources. */
538	RTCritSectDelete(&pAioMgr->CritSectBlockingEvent);
539	RTSemEventDestroy(pAioMgr->EventSem);
540	if (pAioMgr->enmMgrType != PDMACEPFILEMGRTYPE_SIMPLE)
541	pdmacFileAioMgrNormalDestroy(pAioMgr);
542
543	MMR3HeapFree(pAioMgr);
544	}
545
546	static int pdmacFileMgrTypeFromName(const char *pszVal, PPDMACEPFILEMGRTYPE penmMgrType)
547	{
548	int rc = VINF_SUCCESS;
549
550	if (!RTStrCmp(pszVal, "Simple"))
551	*penmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
552	else if (!RTStrCmp(pszVal, "Async"))
553	*penmMgrType = PDMACEPFILEMGRTYPE_ASYNC;
554	else
555	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
556
557	return rc;
558	}
559
560	static const char *pdmacFileMgrTypeToName(PDMACEPFILEMGRTYPE enmMgrType)
561	{
562	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
563	return "Simple";
564	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
565	return "Async";
566
567	return NULL;
568	}
569
570	static int pdmacFileBackendTypeFromName(const char *pszVal, PPDMACFILEEPBACKEND penmBackendType)
571	{
572	int rc = VINF_SUCCESS;
573
574	if (!RTStrCmp(pszVal, "Buffered"))
575	*penmBackendType = PDMACFILEEPBACKEND_BUFFERED;
576	else if (!RTStrCmp(pszVal, "NonBuffered"))
577	*penmBackendType = PDMACFILEEPBACKEND_NON_BUFFERED;
578	else
579	rc = VERR_CFGM_CONFIG_UNKNOWN_VALUE;
580
581	return rc;
582	}
583
584	static const char *pdmacFileBackendTypeToName(PDMACFILEEPBACKEND enmBackendType)
585	{
586	if (enmBackendType == PDMACFILEEPBACKEND_BUFFERED)
587	return "Buffered";
588	if (enmBackendType == PDMACFILEEPBACKEND_NON_BUFFERED)
589	return "NonBuffered";
590
591	return NULL;
592	}
593
594	#ifdef VBOX_WITH_DEBUGGER
595
596	/**
597	* @callback_method_impl{FNDBGCCMD, The '.injecterror' command.}
598	*/
599	static DECLCALLBACK(int) pdmacEpFileErrorInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
600	{
601	/*
602	* Validate input.
603	*/
604	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
605	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 3);
606	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
607	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
608	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
609
610	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
611	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
612
613	/* Syntax is "read\|write <filename> <status code>" */
614	bool fWrite;
615	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
616	fWrite = false;
617	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
618	fWrite = true;
619	else
620	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
621
622	int32_t rcToInject = (int32_t)pArgs[2].u.u64Number;
623	if ((uint64_t)rcToInject != pArgs[2].u.u64Number)
624	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The status code '%lld' is out of range", pArgs[0].u.u64Number);
625
626	/*
627	* Search for the matching endpoint.
628	*/
629	RTCritSectEnter(&pEpClassFile->Core.CritSect);
630
631	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
632	while (pEpFile)
633	{
634	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
635	break;
636	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
637	}
638
639	if (pEpFile)
640	{
641	/*
642	* Do the job.
643	*/
644	if (fWrite)
645	ASMAtomicXchgS32(&pEpFile->rcReqWrite, rcToInject);
646	else
647	ASMAtomicXchgS32(&pEpFile->rcReqRead, rcToInject);
648
649	DBGCCmdHlpPrintf(pCmdHlp, "Injected %Rrc into '%s' for %s\n",
650	(int)rcToInject, pArgs[1].u.pszString, pArgs[0].u.pszString);
651	}
652
653	RTCritSectLeave(&pEpClassFile->Core.CritSect);
654
655	if (!pEpFile)
656	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
657	return VINF_SUCCESS;
658	}
659
660	# ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
661	/**
662	* @callback_method_impl{FNDBGCCMD, The '.injectdelay' command.}
663	*/
664	static DECLCALLBACK(int) pdmacEpFileDelayInject(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PUVM pUVM, PCDBGCVAR pArgs, unsigned cArgs)
665	{
666	/*
667	* Validate input.
668	*/
669	DBGC_CMDHLP_REQ_UVM_RET(pCmdHlp, pCmd, pUVM);
670	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs >= 3);
671	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 0, pArgs[0].enmType == DBGCVAR_TYPE_STRING);
672	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 1, pArgs[1].enmType == DBGCVAR_TYPE_STRING);
673	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, 2, pArgs[2].enmType == DBGCVAR_TYPE_NUMBER);
674
675	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile;
676	pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pUVM->pdm.s.apAsyncCompletionEndpointClass[PDMASYNCCOMPLETIONEPCLASSTYPE_FILE];
677
678	/* Syntax is "read\|write\|flush\|any <filename> <delay> [reqs]" */
679	PDMACFILEREQTYPEDELAY enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
680	if (!RTStrCmp(pArgs[0].u.pszString, "read"))
681	enmDelayType = PDMACFILEREQTYPEDELAY_READ;
682	else if (!RTStrCmp(pArgs[0].u.pszString, "write"))
683	enmDelayType = PDMACFILEREQTYPEDELAY_WRITE;
684	else if (!RTStrCmp(pArgs[0].u.pszString, "flush"))
685	enmDelayType = PDMACFILEREQTYPEDELAY_FLUSH;
686	else if (!RTStrCmp(pArgs[0].u.pszString, "any"))
687	enmDelayType = PDMACFILEREQTYPEDELAY_ANY;
688	else
689	return DBGCCmdHlpFail(pCmdHlp, pCmd, "invalid transfer direction '%s'", pArgs[0].u.pszString);
690
691	uint32_t msDelay = (uint32_t)pArgs[2].u.u64Number;
692	if ((uint64_t)msDelay != pArgs[2].u.u64Number)
693	return DBGCCmdHlpFail(pCmdHlp, pCmd, "The delay '%lld' is out of range", pArgs[0].u.u64Number);
694
695	uint32_t cReqsDelay = 1;
696	uint32_t msJitter = 0;
697	if (cArgs >= 4)
698	msJitter = (uint32_t)pArgs[3].u.u64Number;
699	if (cArgs == 5)
700	cReqsDelay = (uint32_t)pArgs[4].u.u64Number;
701
702	/*
703	* Search for the matching endpoint.
704	*/
705	RTCritSectEnter(&pEpClassFile->Core.CritSect);
706
707	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
708	while (pEpFile)
709	{
710	if (!RTStrCmp(pArgs[1].u.pszString, RTPathFilename(pEpFile->Core.pszUri)))
711	break;
712	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
713	}
714
715	if (pEpFile)
716	{
717	ASMAtomicWriteSize(&pEpFile->enmTypeDelay, enmDelayType);
718	ASMAtomicWriteU32(&pEpFile->msDelay, msDelay);
719	ASMAtomicWriteU32(&pEpFile->msJitter, msJitter);
720	ASMAtomicWriteU32(&pEpFile->cReqsDelay, cReqsDelay);
721
722	DBGCCmdHlpPrintf(pCmdHlp, "Injected delay for the next %u requests of %u ms into '%s' for %s\n",
723	cReqsDelay, msDelay, pArgs[1].u.pszString, pArgs[0].u.pszString);
724	}
725
726	RTCritSectLeave(&pEpClassFile->Core.CritSect);
727
728	if (!pEpFile)
729	return DBGCCmdHlpFail(pCmdHlp, pCmd, "No file with name '%s' found", pArgs[1].u.pszString);
730	return VINF_SUCCESS;
731	}
732
733	static DECLCALLBACK(void) pdmacR3TimerCallback(PVM pVM, PTMTIMER pTimer, void *pvUser)
734	{
735	uint64_t tsCur = RTTimeProgramMilliTS();
736	uint64_t cMilliesNext = UINT64_MAX;
737	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pvUser;
738
739	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, UINT64_MAX);
740
741	/* Go through all endpoints and check for expired requests. */
742	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpClassFile->Core.pEndpointsHead;
743
744	while (pEpFile)
745	{
746	/* Check for an expired delay. */
747	if (pEpFile->pDelayedHead != NULL)
748	{
749	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = ASMAtomicXchgPtrT(&pEpFile->pDelayedHead, NULL, PPDMASYNCCOMPLETIONTASKFILE);
750
751	while (pTaskFile)
752	{
753	PPDMASYNCCOMPLETIONTASKFILE pTmp = pTaskFile;
754	pTaskFile = pTaskFile->pDelayedNext;
755
756	if (tsCur >= pTmp->tsDelayEnd)
757	{
758	LogRel(("AIOMgr: Delayed request %#p completed\n", pTmp));
759	pdmR3AsyncCompletionCompleteTask(&pTmp->Core, pTmp->rc, true);
760	}
761	else
762	{
763	/* Prepend to the delayed list again. */
764	PPDMASYNCCOMPLETIONTASKFILE pHead = NULL;
765
766	if (pTmp->tsDelayEnd - tsCur < cMilliesNext)
767	cMilliesNext = pTmp->tsDelayEnd - tsCur;
768
769	do
770	{
771	pHead = ASMAtomicReadPtrT(&pEpFile->pDelayedHead, PPDMASYNCCOMPLETIONTASKFILE);
772	pTmp->pDelayedNext = pHead;
773	} while (!ASMAtomicCmpXchgPtr(&pEpFile->pDelayedHead, pTmp, pHead));
774	}
775	}
776	}
777
778	pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEpFile->Core.pNext;
779	}
780
781	if (cMilliesNext < pEpClassFile->cMilliesNext)
782	{
783	ASMAtomicWriteU64(&pEpClassFile->cMilliesNext, cMilliesNext);
784	TMTimerSetMillies(pEpClassFile->pTimer, cMilliesNext);
785	}
786	}
787
788	# endif /* PDM_ASYNC_COMPLETION_FILE_WITH_DELAY */
789
790	#endif /* VBOX_WITH_DEBUGGER */
791
792	static int pdmacFileInitialize(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals, PCFGMNODE pCfgNode)
793	{
794	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
795	RTFILEAIOLIMITS AioLimits; /** < Async I/O limitations. */
796
797	int rc = RTFileAioGetLimits(&AioLimits);
798	#ifdef DEBUG
799	if (RT_SUCCESS(rc) && RTEnvExist("VBOX_ASYNC_IO_FAILBACK"))
800	rc = VERR_ENV_VAR_NOT_FOUND;
801	#endif
802	if (RT_FAILURE(rc))
803	{
804	LogRel(("AIO: Async I/O manager not supported (rc=%Rrc). Falling back to simple manager\n", rc));
805	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_SIMPLE;
806	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_BUFFERED;
807	}
808	else
809	{
810	pEpClassFile->uBitmaskAlignment = AioLimits.cbBufferAlignment ? ~((RTR3UINTPTR)AioLimits.cbBufferAlignment - 1) : RTR3UINTPTR_MAX;
811	pEpClassFile->cReqsOutstandingMax = AioLimits.cReqsOutstandingMax;
812
813	if (pCfgNode)
814	{
815	/* Query the default manager type */
816	char *pszVal = NULL;
817	rc = CFGMR3QueryStringAllocDef(pCfgNode, "IoMgr", &pszVal, "Async");
818	AssertLogRelRCReturn(rc, rc);
819
820	rc = pdmacFileMgrTypeFromName(pszVal, &pEpClassFile->enmMgrTypeOverride);
821	MMR3HeapFree(pszVal);
822	if (RT_FAILURE(rc))
823	return rc;
824
825	LogRel(("AIOMgr: Default manager type is '%s'\n", pdmacFileMgrTypeToName(pEpClassFile->enmMgrTypeOverride)));
826
827	/* Query default backend type */
828	rc = CFGMR3QueryStringAllocDef(pCfgNode, "FileBackend", &pszVal, "NonBuffered");
829	AssertLogRelRCReturn(rc, rc);
830
831	rc = pdmacFileBackendTypeFromName(pszVal, &pEpClassFile->enmEpBackendDefault);
832	MMR3HeapFree(pszVal);
833	if (RT_FAILURE(rc))
834	return rc;
835
836	LogRel(("AIOMgr: Default file backend is '%s'\n", pdmacFileBackendTypeToName(pEpClassFile->enmEpBackendDefault)));
837
838	#ifdef RT_OS_LINUX
839	if ( pEpClassFile->enmMgrTypeOverride == PDMACEPFILEMGRTYPE_ASYNC
840	&& pEpClassFile->enmEpBackendDefault == PDMACFILEEPBACKEND_BUFFERED)
841	{
842	LogRel(("AIOMgr: Linux does not support buffered async I/O, changing to non buffered\n"));
843	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
844	}
845	#endif
846	}
847	else
848	{
849	/* No configuration supplied, set defaults */
850	pEpClassFile->enmEpBackendDefault = PDMACFILEEPBACKEND_NON_BUFFERED;
851	pEpClassFile->enmMgrTypeOverride = PDMACEPFILEMGRTYPE_ASYNC;
852	}
853	}
854
855	/* Init critical section. */
856	rc = RTCritSectInit(&pEpClassFile->CritSect);
857
858	#ifdef VBOX_WITH_DEBUGGER
859	/* Install the error injection handler. */
860	if (RT_SUCCESS(rc))
861	{
862	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
863	AssertRC(rc);
864	}
865
866	#ifdef PDM_ASYNC_COMPLETION_FILE_WITH_DELAY
867	rc = TMR3TimerCreateInternal(pEpClassFile->Core.pVM, TMCLOCK_REAL, pdmacR3TimerCallback, pEpClassFile, "AC Delay", &pEpClassFile->pTimer);
868	AssertRC(rc);
869	pEpClassFile->cMilliesNext = UINT64_MAX;
870	#endif
871	#endif
872
873	return rc;
874	}
875
876	static void pdmacFileTerminate(PPDMASYNCCOMPLETIONEPCLASS pClassGlobals)
877	{
878	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pClassGlobals;
879
880	/* All endpoints should be closed at this point. */
881	AssertMsg(!pEpClassFile->Core.pEndpointsHead, ("There are still endpoints left\n"));
882
883	/* Destroy all left async I/O managers. */
884	while (pEpClassFile->pAioMgrHead)
885	pdmacFileAioMgrDestroy(pEpClassFile, pEpClassFile->pAioMgrHead);
886
887	RTCritSectDelete(&pEpClassFile->CritSect);
888	}
889
890	static int pdmacFileEpInitialize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint,
891	const char *pszUri, uint32_t fFlags)
892	{
893	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
894	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
895	PDMACEPFILEMGRTYPE enmMgrType = pEpClassFile->enmMgrTypeOverride;
896	PDMACFILEEPBACKEND enmEpBackend = pEpClassFile->enmEpBackendDefault;
897
898	AssertMsgReturn((fFlags & ~(PDMACEP_FILE_FLAGS_READ_ONLY \| PDMACEP_FILE_FLAGS_DONT_LOCK \| PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)) == 0,
899	("PDMAsyncCompletion: Invalid flag specified\n"), VERR_INVALID_PARAMETER);
900
901	unsigned fFileFlags = RTFILE_O_OPEN;
902
903	/*
904	* Revert to the simple manager and the buffered backend if
905	* the host cache should be enabled.
906	*/
907	if (fFlags & PDMACEP_FILE_FLAGS_HOST_CACHE_ENABLED)
908	{
909	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
910	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
911	}
912
913	if (fFlags & PDMACEP_FILE_FLAGS_READ_ONLY)
914	fFileFlags \|= RTFILE_O_READ \| RTFILE_O_DENY_NONE;
915	else
916	{
917	fFileFlags \|= RTFILE_O_READWRITE;
918
919	/*
920	* Opened in read/write mode. Check whether the caller wants to
921	* avoid the lock. Return an error in case caching is enabled
922	* because this can lead to data corruption.
923	*/
924	if (fFlags & PDMACEP_FILE_FLAGS_DONT_LOCK)
925	fFileFlags \|= RTFILE_O_DENY_NONE;
926	else
927	fFileFlags \|= RTFILE_O_DENY_WRITE;
928	}
929
930	if (enmMgrType == PDMACEPFILEMGRTYPE_ASYNC)
931	fFileFlags \|= RTFILE_O_ASYNC_IO;
932
933	int rc;
934	if (enmEpBackend == PDMACFILEEPBACKEND_NON_BUFFERED)
935	{
936	/*
937	* We only disable the cache if the size of the file is a multiple of 512.
938	* Certain hosts like Windows, Linux and Solaris require that transfer sizes
939	* are aligned to the volume sector size.
940	* If not we just make sure that the data is written to disk with RTFILE_O_WRITE_THROUGH
941	* which will trash the host cache but ensures that the host cache will not
942	* contain dirty buffers.
943	*/
944	RTFILE hFile;
945	rc = RTFileOpen(&hFile, pszUri, RTFILE_O_READ \| RTFILE_O_OPEN \| RTFILE_O_DENY_NONE);
946	if (RT_SUCCESS(rc))
947	{
948	uint64_t cbSize;
949
950	rc = RTFileGetSize(hFile, &cbSize);
951
952	if (RT_SUCCESS(rc) && ((cbSize % 512) == 0))
953	fFileFlags \|= RTFILE_O_NO_CACHE;
954	else
955	{
956	/* Downgrade to the buffered backend */
957	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
958
959	#ifdef RT_OS_LINUX
960	fFileFlags &= ~RTFILE_O_ASYNC_IO;
961	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
962	#endif
963	}
964	RTFileClose(hFile);
965	}
966	}
967
968	/* Open with final flags. */
969	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
970	if ( rc == VERR_INVALID_FUNCTION
971	\|\| rc == VERR_INVALID_PARAMETER)
972	{
973	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed with %Rrc\n",
974	pszUri, fFileFlags, rc));
975	/*
976	* Solaris doesn't support directio on ZFS so far. :-\
977	* Trying to enable it returns VERR_INVALID_FUNCTION
978	* (ENOTTY). Remove it and hope for the best.
979	* ZFS supports write throttling in case applications
980	* write more data than can be synced to the disk
981	* without blocking the whole application.
982	*
983	* On Linux we have the same problem with cifs.
984	* Have to disable async I/O here too because it requires O_DIRECT.
985	*/
986	fFileFlags &= ~RTFILE_O_NO_CACHE;
987	enmEpBackend = PDMACFILEEPBACKEND_BUFFERED;
988
989	#ifdef RT_OS_LINUX
990	fFileFlags &= ~RTFILE_O_ASYNC_IO;
991	enmMgrType = PDMACEPFILEMGRTYPE_SIMPLE;
992	#endif
993
994	/* Open again. */
995	rc = RTFileOpen(&pEpFile->hFile, pszUri, fFileFlags);
996
997	if (RT_FAILURE(rc))
998	{
999	LogRel(("AIOMgr: pdmacFileEpInitialize: RTFileOpen %s / %08x failed AGAIN(!) with %Rrc\n",
1000	pszUri, fFileFlags, rc));
1001	}
1002	}
1003
1004	if (RT_SUCCESS(rc))
1005	{
1006	pEpFile->fFlags = fFileFlags;
1007
1008	rc = RTFileGetSize(pEpFile->hFile, (uint64_t *)&pEpFile->cbFile);
1009	if (RT_SUCCESS(rc))
1010	{
1011	/* Initialize the segment cache */
1012	rc = MMR3HeapAllocZEx(pEpClassFile->Core.pVM, MM_TAG_PDM_ASYNC_COMPLETION,
1013	sizeof(PDMACTASKFILE),
1014	(void **)&pEpFile->pTasksFreeHead);
1015	if (RT_SUCCESS(rc))
1016	{
1017	PPDMACEPFILEMGR pAioMgr = NULL;
1018
1019	pEpFile->pTasksFreeTail = pEpFile->pTasksFreeHead;
1020	pEpFile->cTasksCached = 0;
1021	pEpFile->enmBackendType = enmEpBackend;
1022	/*
1023	* Disable async flushes on Solaris for now.
1024	* They cause weird hangs which needs more investigations.
1025	*/
1026	#ifndef RT_OS_SOLARIS
1027	pEpFile->fAsyncFlushSupported = true;
1028	#else
1029	pEpFile->fAsyncFlushSupported = false;
1030	#endif
1031
1032	if (enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1033	{
1034	/* Simple mode. Every file has its own async I/O manager. */
1035	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, PDMACEPFILEMGRTYPE_SIMPLE);
1036	}
1037	else
1038	{
1039	pAioMgr = pEpClassFile->pAioMgrHead;
1040
1041	/* Check for an idling manager of the same type */
1042	while (pAioMgr)
1043	{
1044	if (pAioMgr->enmMgrType == enmMgrType)
1045	break;
1046	pAioMgr = pAioMgr->pNext;
1047	}
1048
1049	if (!pAioMgr)
1050	rc = pdmacFileAioMgrCreate(pEpClassFile, &pAioMgr, enmMgrType);
1051	}
1052
1053	if (RT_SUCCESS(rc))
1054	{
1055	pEpFile->AioMgr.pTreeRangesLocked = (PAVLRFOFFTREE)RTMemAllocZ(sizeof(AVLRFOFFTREE));
1056	if (!pEpFile->AioMgr.pTreeRangesLocked)
1057	rc = VERR_NO_MEMORY;
1058	else
1059	{
1060	pEpFile->enmState = PDMASYNCCOMPLETIONENDPOINTFILESTATE_ACTIVE;
1061
1062	/* Assign the endpoint to the thread. */
1063	rc = pdmacFileAioMgrAddEndpoint(pAioMgr, pEpFile);
1064	if (RT_FAILURE(rc))
1065	{
1066	RTMemFree(pEpFile->AioMgr.pTreeRangesLocked);
1067	MMR3HeapFree(pEpFile->pTasksFreeHead);
1068	}
1069	}
1070	}
1071	else if (rc == VERR_FILE_AIO_INSUFFICIENT_EVENTS)
1072	{
1073	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1074	#if defined(RT_OS_LINUX)
1075	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1076	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1077	"Either increase the amount of allowed events in /proc/sys/fs/aio-max-nr or enable "
1078	"the host I/O cache"));
1079	#else
1080	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1081	N_("Failed to create I/O manager for VM due to insufficient resources on the host. "
1082	"Enable the host I/O cache"));
1083	#endif
1084	}
1085	else
1086	{
1087	PUVM pUVM = VMR3GetUVM(pEpClassFile->Core.pVM);
1088	rc = VMR3SetError(pUVM, rc, RT_SRC_POS,
1089	N_("Failed to create I/O manager for VM due to an unknown error"));
1090	}
1091	}
1092	}
1093
1094	if (RT_FAILURE(rc))
1095	RTFileClose(pEpFile->hFile);
1096	}
1097
1098	#ifdef VBOX_WITH_STATISTICS
1099	if (RT_SUCCESS(rc))
1100	{
1101	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatRead,
1102	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1103	STAMUNIT_TICKS_PER_CALL, "Time taken to read from the endpoint",
1104	"/PDM/AsyncCompletion/File/%s/Read", RTPathFilename(pEpFile->Core.pszUri));
1105
1106	STAMR3RegisterF(pEpClassFile->Core.pVM, &pEpFile->StatWrite,
1107	STAMTYPE_PROFILE_ADV, STAMVISIBILITY_ALWAYS,
1108	STAMUNIT_TICKS_PER_CALL, "Time taken to write to the endpoint",
1109	"/PDM/AsyncCompletion/File/%s/Write", RTPathFilename(pEpFile->Core.pszUri));
1110	}
1111	#endif
1112
1113	if (RT_SUCCESS(rc))
1114	LogRel(("AIOMgr: Endpoint for file '%s' (flags %08x) created successfully\n", pszUri, pEpFile->fFlags));
1115
1116	return rc;
1117	}
1118
1119	static int pdmacFileEpRangesLockedDestroy(PAVLRFOFFNODECORE pNode, void *pvUser)
1120	{
1121	NOREF(pNode); NOREF(pvUser);
1122	AssertMsgFailed(("The locked ranges tree should be empty at that point\n"));
1123	return VINF_SUCCESS;
1124	}
1125
1126	static int pdmacFileEpClose(PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1127	{
1128	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1129	PPDMASYNCCOMPLETIONEPCLASSFILE pEpClassFile = (PPDMASYNCCOMPLETIONEPCLASSFILE)pEndpoint->pEpClass;
1130
1131	/* Make sure that all tasks finished for this endpoint. */
1132	int rc = pdmacFileAioMgrCloseEndpoint(pEpFile->pAioMgr, pEpFile);
1133	AssertRC(rc);
1134
1135	/*
1136	* If the async I/O manager is in failsafe mode this is the only endpoint
1137	* he processes and thus can be destroyed now.
1138	*/
1139	if (pEpFile->pAioMgr->enmMgrType == PDMACEPFILEMGRTYPE_SIMPLE)
1140	pdmacFileAioMgrDestroy(pEpClassFile, pEpFile->pAioMgr);
1141
1142	/* Free cached tasks. */
1143	PPDMACTASKFILE pTask = pEpFile->pTasksFreeHead;
1144
1145	while (pTask)
1146	{
1147	PPDMACTASKFILE pTaskFree = pTask;
1148	pTask = pTask->pNext;
1149	MMR3HeapFree(pTaskFree);
1150	}
1151
1152	/* Destroy the locked ranges tree now. */
1153	RTAvlrFileOffsetDestroy(pEpFile->AioMgr.pTreeRangesLocked, pdmacFileEpRangesLockedDestroy, NULL);
1154
1155	RTFileClose(pEpFile->hFile);
1156
1157	#ifdef VBOX_WITH_STATISTICS
1158	/* Not sure if this might be unnecessary because of similar statement in pdmR3AsyncCompletionStatisticsDeregister? */
1159	STAMR3DeregisterF(pEpClassFile->Core.pVM->pUVM, "/PDM/AsyncCompletion/File/%s/*", RTPathFilename(pEpFile->Core.pszUri));
1160	#endif
1161
1162	return VINF_SUCCESS;
1163	}
1164
1165	static int pdmacFileEpRead(PPDMASYNCCOMPLETIONTASK pTask,
1166	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1167	PCRTSGSEG paSegments, size_t cSegments,
1168	size_t cbRead)
1169	{
1170	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1171
1172	LogFlowFunc(("pTask=%#p pEndpoint=%#p off=%RTfoff paSegments=%#p cSegments=%zu cbRead=%zu\n",
1173	pTask, pEndpoint, off, paSegments, cSegments, cbRead));
1174
1175	if (RT_UNLIKELY((uint64_t)off + cbRead > pEpFile->cbFile))
1176	return VERR_EOF;
1177
1178	STAM_PROFILE_ADV_START(&pEpFile->StatRead, Read);
1179	pdmacFileEpTaskInit(pTask, cbRead);
1180	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbRead,
1181	PDMACTASKFILETRANSFER_READ);
1182	STAM_PROFILE_ADV_STOP(&pEpFile->StatRead, Read);
1183
1184	return rc;
1185	}
1186
1187	static int pdmacFileEpWrite(PPDMASYNCCOMPLETIONTASK pTask,
1188	PPDMASYNCCOMPLETIONENDPOINT pEndpoint, RTFOFF off,
1189	PCRTSGSEG paSegments, size_t cSegments,
1190	size_t cbWrite)
1191	{
1192	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1193
1194	if (RT_UNLIKELY(pEpFile->fReadonly))
1195	return VERR_NOT_SUPPORTED;
1196
1197	STAM_PROFILE_ADV_START(&pEpFile->StatWrite, Write);
1198
1199	pdmacFileEpTaskInit(pTask, cbWrite);
1200
1201	int rc = pdmacFileEpTaskInitiate(pTask, pEndpoint, off, paSegments, cSegments, cbWrite,
1202	PDMACTASKFILETRANSFER_WRITE);
1203
1204	STAM_PROFILE_ADV_STOP(&pEpFile->StatWrite, Write);
1205
1206	return rc;
1207	}
1208
1209	static int pdmacFileEpFlush(PPDMASYNCCOMPLETIONTASK pTask,
1210	PPDMASYNCCOMPLETIONENDPOINT pEndpoint)
1211	{
1212	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1213	PPDMASYNCCOMPLETIONTASKFILE pTaskFile = (PPDMASYNCCOMPLETIONTASKFILE)pTask;
1214
1215	if (RT_UNLIKELY(pEpFile->fReadonly))
1216	return VERR_NOT_SUPPORTED;
1217
1218	pdmacFileEpTaskInit(pTask, 0);
1219
1220	PPDMACTASKFILE pIoTask = pdmacFileTaskAlloc(pEpFile);
1221	if (RT_UNLIKELY(!pIoTask))
1222	return VERR_NO_MEMORY;
1223
1224	pIoTask->pEndpoint = pEpFile;
1225	pIoTask->enmTransferType = PDMACTASKFILETRANSFER_FLUSH;
1226	pIoTask->pvUser = pTaskFile;
1227	pIoTask->pfnCompleted = pdmacFileEpTaskCompleted;
1228	pdmacFileEpAddTask(pEpFile, pIoTask);
1229
1230	return VINF_AIO_TASK_PENDING;
1231	}
1232
1233	static int pdmacFileEpGetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t *pcbSize)
1234	{
1235	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1236
1237	*pcbSize = ASMAtomicReadU64(&pEpFile->cbFile);
1238
1239	return VINF_SUCCESS;
1240	}
1241
1242	static int pdmacFileEpSetSize(PPDMASYNCCOMPLETIONENDPOINT pEndpoint, uint64_t cbSize)
1243	{
1244	int rc;
1245	PPDMASYNCCOMPLETIONENDPOINTFILE pEpFile = (PPDMASYNCCOMPLETIONENDPOINTFILE)pEndpoint;
1246
1247	rc = RTFileSetSize(pEpFile->hFile, cbSize);
1248	if (RT_SUCCESS(rc))
1249	ASMAtomicWriteU64(&pEpFile->cbFile, cbSize);
1250
1251	return rc;
1252	}
1253
1254	const PDMASYNCCOMPLETIONEPCLASSOPS g_PDMAsyncCompletionEndpointClassFile =
1255	{
1256	/* u32Version */
1257	PDMAC_EPCLASS_OPS_VERSION,
1258	/* pcszName */
1259	"File",
1260	/* enmClassType */
1261	PDMASYNCCOMPLETIONEPCLASSTYPE_FILE,
1262	/* cbEndpointClassGlobal */
1263	sizeof(PDMASYNCCOMPLETIONEPCLASSFILE),
1264	/* cbEndpoint */
1265	sizeof(PDMASYNCCOMPLETIONENDPOINTFILE),
1266	/* cbTask */
1267	sizeof(PDMASYNCCOMPLETIONTASKFILE),
1268	/* pfnInitialize */
1269	pdmacFileInitialize,
1270	/* pfnTerminate */
1271	pdmacFileTerminate,
1272	/* pfnEpInitialize. */
1273	pdmacFileEpInitialize,
1274	/* pfnEpClose */
1275	pdmacFileEpClose,
1276	/* pfnEpRead */
1277	pdmacFileEpRead,
1278	/* pfnEpWrite */
1279	pdmacFileEpWrite,
1280	/* pfnEpFlush */
1281	pdmacFileEpFlush,
1282	/* pfnEpGetSize */
1283	pdmacFileEpGetSize,
1284	/* pfnEpSetSize */
1285	pdmacFileEpSetSize,
1286	/* u32VersionEnd */
1287	PDMAC_EPCLASS_OPS_VERSION
1288	};
1289

注意: 瀏覽 TracBrowser 來幫助您使用儲存庫瀏覽器

source: vbox/trunk/src/VBox/VMM/VMMR3/PDMAsyncCompletionFile.cpp@ 57358

以其他格式下載: